Table 1 RMSEs of the flood dataset computed using the proposed framework, randomly distributed embedding (RDE), and multiview embedding (MVE) and also compared with those obtained by long short-term memory (LSTM), support vector regression (SVR), and the random forest.

From: Forecasting high-dimensional dynamics exploiting suboptimal embeddings

Prediction steps [h]

Proposed

RDE

MVE

LSTM

SVR

Random Forest

6

0.068

0.069

0.060

0.123

0.073

0.079

12

0.167

0.174

0.171

0.213

0.177

0.185

18

0.247

0.278

0.273

0.291

0.268

0.275

24

0.349

0.389

0.373

0.372

0.365

0.363

  1. The proposed framework achieved the best accuracy for the 12-, 18-, and 24-h-ahead forecasts. Although the result obtained by the proposed framework for the 6-h-ahead forecast is the second best, the difference is slight.
Back to article page